We have developed a method of long-term EMG recording that has proven suitable for the quantification of pathological tremor. In the present paper we show that the principal parameters of the method (tremor occurrence, tremor intensity, tremor frequency) are highly reproducible and that the method is specific and sensitive for detection of pathological tremor. Twelve patients with essential tremor (ET) and 13 patients with Parkinson's disease (PD) were recorded repeatedly on 3 successive days. For each patient group and for each parameter the intersubject variability was much larger than the intersubject variability. The intraclass correlation coefficient "R" was in the order of 0.9 for each parameter and the mean of Pearson's correlation coefficient between successive days was also approximately 0.9. Recordings from normal controls demonstrated that the method's specificity for pathological tremor is 94.1% and its sensitivity is 96%.